AI And Machine Learning In Healthcare
There’s a lot to say about how artificial intelligence will disrupt healthcare as we know it. Here’s a crash course in what AI and machine learning mean for healthcare today and what the future could look like for these technologies.
Algorithmic Diagnosis, No Doctor Required
In 2018, the U.S. FDA approved an industry first: they gave the go-ahead to begin marketing an artificial intelligence platform that can automatically detect mild and moderate cases of diabetic retinopathy. This condition causes vision loss in a significant number of the 30 million Americanswho have diabetes. As with many other conditions, early detection is key. That’s where AI comes in.
Using a retinal camera, patients have images of their eyes taken and then uploaded to servers with the compute power to run this newly FDA-approved software package, called “IDx-DR.” If the images are of high enough quality, it will be able to identify patients with indications of diabetic retinopathy early enough in development that they’ll be able to seek aggressive treatment with the help of a specialist.
The distinction which makes this “device” a first-in-class product is that it doesn’t require a human clinician to interpret patient records (retinal photographs, in this case) and arrive at a diagnosis. What this means is that doctors of any background could soon deliver a test like this one — or any other — even if they’re not primarily eye doctors themselves.
As for the results, the FDA reports that IDx-DR correctly detected the presence of mild cases of diabetic retinopathy in 87.4% of test cases — a factor which surely helped make the decision to grant marketing approval easier.
This is far from the only example of machine learning in diagnostic medicine. Researchers at the University of Queensland in Australia used audio recordings gathered from almost 1,500 pediatric patients, spanning the ages of just days old all the way to age 12, to “train” a software application to diagnose respiratory tract disease, bronchiolitis, croup, asthma and pneumonia with an accuracy of 81% to 97%. It’s literally an app that makes diagnoses based on the way kids cough, and it’s something our smartphones could run.
Examples like these are extraordinary. They show that geography will soon be only a trivial barrier when it comes to healthcare, at least where making patient connections and rendering timely diagnoses are concerned. That’s a huge step forward in meeting the needs of underserved communities. Patients with mobility issues will be able to receive appraisals of their conditions and their progress, even if they can’t travel or their regular care team is unavailable.
Generative Design for Medication Discovery
The always-improving processing power of our computers over the years has ensured that the design complexity of our products has improved just as reliably. But thanks to machine learning, product designers in a variety of fields — including healthcare, furniture making, architecture, transportation, aviation and many others — are turning to generative design to envision and test a greater variety of products than ever before, in order to find the design that most effectively solves a known problem. In generative design, AI comes up with prototypes that satisfy basic requirements for weight, material costs, structural integrity, and other variables.
In medicine, the AI “tests for” a different variable: the ability to treat a symptom or known condition. Generative design in healthcare involves using neural networks for the task of testing many different permutations all at once, with the goal of coming up with entirely new chemical formulations for pharmaceuticals. The result could be an explosion in useful products, and a true unlocking of the pharmaceutical industry’s potential, which used to be limited by trial and error (and trial and error’s hefty financial price tag).
Data Pooling for Treatment Personalisation and Predicting Patient Outcomes
There’s another application for cloud-based artificial intelligence that’s closely related to the testing of new chemical formulations and potential pharma products. It’s the personalisation of patient treatment in other ways — including surgeries and therapies, hormone treatments, wearable medical devices, and more.
Some technology-driven healthcare companies offer tools for safely and securely pooling patient data from multiple facilities and participating institutions. This pooling of data in the cloud permits large-scale analysis for the development of new approaches to treatment. In other words, a larger set of patient experiences means a better chance of zeroing in on, and then perfecting, those treatments which work with patients matching a particular profile.
Researchers have also long been interested in ways to predict patient outcomes and the likelihood of specific treatments working instead of others. Researchers from Cornell University proposed their own method for using deep learning, plus de-identified electronic health records (EHRs), to predict things like mortality rates in hospitals, patient readmissions, lengths of stay and more. According to the researchers, this neural network actually outperforms existing predictive models used to make decisions in clinical settings.
Consider the scale of that achievement for a moment. This is a tool that just in its trial runs leveraged data from more than 216,000 patients, for a total of more than 46 billion discrete data points — including data points which originated in written clinical notes. As artificial intelligence applications like this one grow, they’ll have even more data to draw from, meaning the recommendations will become more accurate as well as more scalable over time.
The Future of Machine Learning in Healthcare
The healthcare market, like many other industries, is waking to the huge potential of artificial intelligence, machine learning, and neural networks. Unfortunately, this also means every industry is also experiencing and contributing to the same ongoing talent shortage in fields like data science, network architecture, software development, computer vision, AI, and several other fields. The average company will begin feeling the competitive pinch if they don’t take on high-level technology hires by 2020, some experts say.
Even if staffing is to remain a pain point for the foreseeable future, that shouldn’t dull our optimism when it comes to AI’s potential in healthcare. This is a groundbreaking set of tools that will help improve patient outcomes across the world.